Different results with new release Mplus PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
Message/Author
 Aurelie Lange posted on Monday, October 10, 2016 - 3:13 am
Dear Dr Muthén and Muthén,

I recently downloaded Mplus 7.4. Based on comments of reviewers I requested some additional output for analyses I conducted with version 7.3. However, I get slightly different results. Most of the time, estimates differ in the second or third decimal, but for one of the analyses it has changed the interpretation of the results.
Comparing the outputs I have come to realise version 7.4 uses more integration points than 7.3. Specifying the integration points in the input file does make the estimates with 7.4 more similar to the output with version 7.3, but not completely, so I suppose there must be some other minor differences between both versions.

Are the results from 7.4 more accurate and should I report those? Or can I change the number of integration points to mimic as much as possible the output I got with 7.3?

Thanks!
Aurelie
 Bengt O. Muthen posted on Monday, October 10, 2016 - 8:55 am
In principle you should go with the solution that has the best loglikelihood. But if you have high number of dimensions of integration, the likelihood may be poorly determined. So to answer your question I need to know how many dimensions you have, how many integration points were used in your 7.3 run, and how many in your 7.4 run.
 Aurelie Lange posted on Monday, October 10, 2016 - 11:10 am
Dear Dr Muthen,

The largest differences between version 7.3 and 7.4 are found for a complex twolevel model with 3 binary outcome variables on the within and between level.

With version 7.3:
Loglikelihood: -12083.147
Dimensions of integration: 5
Number of integration points: 426

With version 7.4
Loglikelihood; -12078.203
Dimensions of integration: 5
Number of integration points: 4250

Thank you for your help.

Sincerely,
Aurelie
 Bengt O. Muthen posted on Monday, October 10, 2016 - 2:49 pm
You should then go with the 7.4 results because this solution has more precision due to more integration points and has a better logL.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: